Defect Image Classification Apparatus Using Switching Logs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, the efficiency of manual defect classification (MDC) is hindered by the need for manual confirmation and modification of automatic defect classification (ADC) results, as ADC cannot completely classify a wide variety of defects, leading to inefficiencies in defect image classification.
Innovation Solution
A defect image classification apparatus that includes an image acquisition unit, automatic classification processing, manual classification processing, and a control unit to selectively display images from multiple detectors based on user input and switching logs, optimizing the initial display for defect kind specification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If ADC is used for defect classification, then classification speed is improved, but classification accuracy deteriorates for wide variety of defects
Solution Approach 1:
The patent segments the classification workflow into two distinct phases: automatic classification (ADC) for initial high-speed processing, and manual classification (MDC) for accurate verification of ambiguous cases. This segmentation allows the system to leverage the speed of ADC while maintaining the accuracy of MDC through selective application.
Solution Approach 2:
The patent applies partial automation by using ADC only for the portion of defects that can be confidently classified, while reserving MDC for cases requiring higher accuracy. The system performs excessive classification actions (both ADC and MDC) on borderline cases to ensure accuracy, rather than relying solely on ADC.
2Measurement precision
If MDC is performed to improve classification accuracy, then classification accuracy is improved, but work efficiency deteriorates
Solution Approach 1:
The patent performs preliminary classification using ADC before MDC. By pre-processing defect images through automatic classification, the system reduces the volume of images requiring manual review, thereby maintaining accuracy while improving overall work efficiency.
Solution Approach 2:
The system implements feedback mechanisms where MDC results are used to refine and improve ADC performance over time. This feedback loop allows the automatic classifier to learn from manual corrections, progressively improving accuracy while reducing the burden of manual work.
3Measurement precision
If multiple defect images are displayed for comparison in MDC, then defect specification accuracy is improved, but operation complexity increases
Solution Approach 1:
The patent extracts and displays only the most relevant defect images for comparison, rather than showing all available images. By selectively extracting key images that provide the most diagnostic value, the system maintains specification accuracy while reducing operational complexity.
Solution Approach 2:
The system applies local quality by displaying different numbers of images depending on the specific defect type and classification context. For well-defined defects, fewer images are displayed, while for ambiguous cases, more images are shown to aid specification, optimizing the balance between accuracy and complexity.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Improves the efficiency of MDC by ensuring that the most suitable images for defect kind specification are displayed initially, reducing the need for manual switching and enhancing the accuracy and speed of defect classification.
Implementation Method 1
signals obtained by detecting secondary particles obtained by emitting charged particle beams to a sample by the plurality of detectors
Data Source
AI summary
A defect image classification apparatus includes a control unit that selects images obtained from at least some detectors among a plurality of detectors, associated with kinds of defects to be a classification result of an automatic defect classification processing unit, as images displayed initially on a display unit. The control unit associates the kinds of the defects and the images displayed initially on the display unit, on the basis of a switching operation log when a user classifies images of defects determined previously as the same kinds as the kinds of the defects determined by the automatic defect classification processing unit.